tasks/text.py

import torch
import random
import torch.nn as nn
from fastapi import APIRouter
from datetime import datetime
from datasets import load_dataset
from sklearn.metrics import accuracy_score
from transformers import AutoTokenizer, AutoModel, AutoConfig
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file

from .utils.evaluation import TextEvaluationRequest
from .utils.emissions import tracker, clean_emissions_data, get_space_info

DESCRIPTION = "GTE Architecture"
ROUTE = "/text"

class AutoBertClassifier(nn.Module):
    def __init__(self, num_labels=8, model_path="haisongzhang/roberta-tiny-cased"):
        super().__init__()
        self.tokenizer = AutoTokenizer.from_pretrained(model_path)
        self.bert = AutoModel.from_pretrained(model_path)
        self.config = AutoConfig.from_pretrained(model_path)
        self.config.num_labels = num_labels
        self.dropout = nn.Dropout(0.05)
        self.classifier = nn.Linear(self.bert.config.hidden_size, num_labels)

    def forward(self, input_ids, attention_mask):
        outputs = self.bert(input_ids=input_ids, attention_mask=attention_mask)
        pooled_output = outputs.last_hidden_state[:, 0]  # Using [CLS] token representation
        pooled_output = self.dropout(pooled_output)
        logits = self.classifier(pooled_output)
        return logits

device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
    
model_repo = "elucidator8918/frugal-ai-text-tiny-final"
model = AutoBertClassifier(num_labels=8)
model.load_state_dict(load_file(hf_hub_download(repo_id=model_repo, filename="model.safetensors")))
tokenizer = AutoTokenizer.from_pretrained(model_repo)

model = model.to(device)
model.eval()

router = APIRouter()

@router.post(ROUTE, tags=["Text Task"], 
             description=DESCRIPTION)
async def evaluate_text(request: TextEvaluationRequest):
    """
    Evaluate text classification for climate disinformation detection.
    
    Current Model: GTE Architecture
    """
    # Get space info
    username, space_url = get_space_info()

    # Define the label mapping
    LABEL_MAPPING = {
        "0_not_relevant": 0,
        "1_not_happening": 1,
        "2_not_human": 2,
        "3_not_bad": 3,
        "4_solutions_harmful_unnecessary": 4,
        "5_science_unreliable": 5,
        "6_proponents_biased": 6,
        "7_fossil_fuels_needed": 7
    }

    # Load and prepare the dataset
    dataset = load_dataset(request.dataset_name)

    # Convert string labels to integers
    dataset = dataset.map(lambda x: {"label": LABEL_MAPPING[x["label"]]})

    # Split dataset
    train_test = dataset["train"].train_test_split(test_size=request.test_size, seed=request.test_seed)
    test_dataset = train_test["test"]

    true_labels = test_dataset["label"]
    texts = test_dataset["quote"]
    
    # Start tracking emissions
    tracker.start()
    tracker.start_task("inference")

    #--------------------------------------------------------------------------------------------
    # YOUR MODEL INFERENCE CODE HERE
    # Update the code below to replace the random baseline by your model inference within the inference pass where the energy consumption and emissions are tracked.
    #--------------------------------------------------------------------------------------------   
    
    text_encoding = tokenizer(
        texts,
        truncation=True,
        padding=True,
        return_tensors="pt",
        max_length=256
    )

    with torch.no_grad():
        text_input_ids = text_encoding["input_ids"].to(device)
        text_attention_mask = text_encoding["attention_mask"].to(device)
        logits = model(text_input_ids, text_attention_mask)
        predictions = torch.argmax(logits, dim=1).cpu().numpy()
    
    #--------------------------------------------------------------------------------------------
    # YOUR MODEL INFERENCE STOPS HERE
    #--------------------------------------------------------------------------------------------   

    # Stop tracking emissions
    emissions_data = tracker.stop_task()
    
    # Calculate accuracy
    accuracy = accuracy_score(true_labels, predictions)

    print(f"Accuracy = {accuracy}")
    
    # Prepare results dictionary
    results = {
        "username": username,
        "space_url": space_url,
        "submission_timestamp": datetime.now().isoformat(),
        "model_description": DESCRIPTION,
        "accuracy": float(accuracy),
        "energy_consumed_wh": emissions_data.energy_consumed * 1000,
        "emissions_gco2eq": emissions_data.emissions * 1000,
        "emissions_data": clean_emissions_data(emissions_data),
        "api_route": ROUTE,
        "dataset_config": {
            "dataset_name": request.dataset_name,
            "test_size": request.test_size,
            "test_seed": request.test_seed
        }
    }

    print(results)
    
    return results